Application of local climate zones combined with machine learning to predict the impact of urban structure patterns on thermal environment

The urban heat island effect is a pertinent concern and estimating the thermal stress in cities is a crucial research goal. Most relevant studies have focused on the relationship between urban structures and high air temperatures, ignoring the effect of different combination of urban structures on high air temperature identifications. In this study, an efficient method was developed for identifying the thermal risk based on combination of urban structures to explore the relationship between urban environment and meteorological conditions. After the preprocessing of LCZ and meteorological data, data were input into the convolutional neural network (CNN) model constructed in this study. After the training process, the accuracy rate of the proposed model increased as the sampling rate increased, and the model achieved a high accuracy rate of 81.97% at 50% sampling. Furthermore, the model performed particularly well at identifying low and medium level of thermal risk. This means that the model could estimate the classes of thermal risk in most areas in Taipei City. The feasibility of using the proposed CNN model to identify the level of thermal risk was proven; the model is a convenient tool for estimating thermal stress in different areas with incomplete air temperature data.